A container image cleared yesterday can become a problem today when a package inside it is identified as malicious. CrowdStrike’s November 4 Falcon Cloud Security release says its new, generally available controls reassess previously scanned images when new package threats are confirmed, find affected running containers, and let a Kubernetes admission policy block a compromised image before deployment. The immediate audience is the team that owns package intake, image registries and Kubernetes release policy.

CrowdStrike says it observed ALTERED SPIDER compromise more than 300 software dependencies in a single day. For a team running Falcon Cloud Security, the release changes how package findings reach production deployment decisions. Check whether a newly flagged package in an older image reaches the admission policy for the next deployment.

Image scans meet admission policy

CrowdStrike says Supply Chain Security for Cloud looks for compromised npm and PyPI packages in images during CI/CD and registry scanning. Its intelligence can reassess images already scanned when a new package threat is confirmed, then identify affected containers already running. The vendor puts the problem plainly: “A compromised dependency does not need a CVE to be dangerous.”

CrowdStrike says the package intelligence it uses to detect and quarantine malicious packages on endpoints now feeds Supply Chain Security for Cloud. An image reassessed after a new package finding can reach the admission policy even when its original CI/CD scan showed no finding.

The admission decision happens later, at deployment. CrowdStrike says the Falcon Kubernetes Admission Controller uses image-assessment policies to alert on or block a compromised image before it reaches production. An alert policy surfaces a finding; a block policy prevents that deployment. Already running containers need their own response, because an admission decision governs a new deployment request. The Falcon Cloud Security product page also describes detection during CI/CD and registry scanning and blocking at Kubernetes admission. Test the configured policy against your own admission decisions.

Cloud detections widen beyond AWS

The release also extends real-time cloud detection and response to Microsoft Azure and Kubernetes, alongside existing AWS and Google Cloud coverage. CrowdStrike says it analyzes streaming cloud audit activity and joins suspicious behavior with identity, asset, workload and exposure context. A privileged action from a dormant identity, for example, receives a different priority when that identity also has elevated permissions or can reach sensitive data. Operators can deny credentials or shut down compromised instances in the Falcon console; CrowdStrike also offers automated actions through Charlotte Agentic SOAR.

AWS anomalies and cloud risk context

AWS CloudTrail gains a separate AI-based anomaly detection model that learns a customer’s baseline across identities, APIs and resources. CrowdStrike says the model evaluates deviations across multiple events to help uncover account takeover and API misuse. The announcement distinguishes that AWS-specific model from the wider Azure and Kubernetes real-time CDR expansion. It also adds Azure and Google Cloud risk context, Google Compute Engine agentless scanning for Linux instances, and more OCI inventory. Those are visibility and prioritization changes, while the image admission policy is the explicit predeployment stop in this release.

Check the image decision in your cluster

Start with the registries and CI/CD pipelines feeding your Kubernetes clusters. Confirm which images Falcon scans, then compare newly flagged image digests with both registry inventory and the running-container matches surfaced by Falcon. Review reassessment findings alongside original build-scan results. Route matches in running workloads to your incident-response team to assess the package, its execution context and the workload’s access before redeploying a clean image.

Inspect the Falcon Kubernetes Admission Controller policy on the clusters that receive those images. Set an appropriate block action for confirmed compromised images; keep an approved staging rollout to measure whether the policy denies deployment. Compare a vendor-flagged image digest with the controller’s recorded decision in staging: the expected result under a block policy is a denied deployment, followed by a separate investigation of any existing container with that digest. If the digest is absent from scan coverage, extend CI or registry assessment before relying on the admission gate.

For real-time cloud detection, confirm Azure and Kubernetes audit events reach the Falcon CDR view where you investigate them. Identify the person or automation authorized to deny credentials or shut down instances, and test that response workflow in a controlled environment.